2004-05-25 7,425 share

Book review: The Cognitive Structure of Emotions

Throughout intellectual history, countless psychologists, linguists, and philosophers have touched upon the subject of emotions, and many have proposed ontologies for them. In The Cognitive Structure of Emotions, Ortony, Clore, and Collins (OC&C) give a particularly lucid discussion of the shortcomings and gaps exhibited in the prominent theories of emotions which have arisen in prior work, and then themselves propose quite an elegant and well-elaborated framework for emotions. Their framework is at once intuitive and predictive, and bears implications for the computation of emotions.

OC&C portray the major shortcoming of prior theories on emotions as discussing more the phenomenology and entailments of emotions rather than describing the cognitive origins of emotions. OC&C make the case for a small ontology of cognitive appraisal processes which constitutes the eliciting conditions for emotions. The cognitive emotions have largely been obscured in prior work. For example, in William James's famous paper What is an emotion? (1884), James introduces surprise, curiosity, rapture, fear, anger, lust, and greed as "The Standard Emotions," whereas OC&C argue that these are simply the low-cognitive, rather reactive, states (and to be precise, the authors don't consider many of these emotions at all). High-cognitive emotions such as shame and embarrassment require more complex cognitive appraisal, and these are the class of emotions which benefit most from OC&C's theory. OC&C characterise prior theories of emotions as falling into two groups: dimensional scaling theories, and arousal-appraisal theories (e.g. Mandler, 1975). The authors enumerate four sources of evidence that prior emotional theories draw from: linguistic, self-reported experiential, behavioural, and physiological evidence; they criticise that often times language, behavior, and physiological patterns have been equated with emotions through one-to-one mappings, and prefer to motivate the study of emotions as cognitively triggered phenomena.

OC&C characterises emotions as valenced reactions to events, agents, or objects, with their particular nature being determined by the way in which the eliciting situation is construed. Their theory of emotions is that it arises out of the cognitive appraisal of an event, a causing agent (animates, or projections onto inanimates), and objects (inanimates, or depersonified animates), additionally remarking that appraisal typically happens in this left-to-right ordering. Although not clearly articulated in the book, there is a temporal logic to this left-to-right evaluation insofar as events are the most immediate (reactive), finding the root cause takes some time (deliberative), and accreting what's felt about the agent onto attitudes about objects only happens after a longer time of reflexive conditioning (sub-liminal reflection). Also, it is important to realise that an event, agent, and object may in fact be differing perspectives of the same thing, and that the only difference is the stance taken toward it. Perhaps it makes sense to map event, agent, and object onto Dan Dennett's physical, intentional, and design stances, respectively. In addition to the articulation of event, agent, and object, OC&C articulate two dimensions of emotion, the first being valence, and the second being intensity. Also implicit in OC&C model is the idea of inwardly appraised emotions (self-reflection) versus outwardly appraised emotions (other-reflection). The diagram on page 19 almost entirely captures the theory.

One of the major implications of the theory is that many states which have previously been called emotions in prior work do not meet OC&C's cognitive-origins definition. For example, surprise and "being abandoned" are not OC&C emotions because they are cognitive states rather than affective states. While "being abandoned" is often the antecedent leading to fear, anxiety, etc, OC&C point out that emotions are not an eventuality of "being abandoned." It is a sensible point in general to tease apart non-emotional antecedent states from the affective consequent states. Furthermore, OC&C point out that affective states may be undifferentiated or differentiated. Usually the typical flow is from unemotional state to undifferentiated affective reaction (e.g. feeling displeased) to differentiated emotions (e.g. resentment, reproach, anger). There is a superpositioning relationship between this flow and the left-to-right appraisal flow. Anger is typically felt not about an event, but about an agent causing the event. Finding the cause gives shape and directedness to emotions, clarifying the vagueries of "events."

In the details of OC&C's theory lie some nice confirmatory parallels to my own work on emotions and attitudes. In my What Would They Think? project (Liu & Maes, 2004), the idea that attitudes are the result of reflexive conditioning is consistent with the notion that objective appraisal involves instaneous, "unthoughtful" attraction judgements, much like the idea that human pre-dispositions like tastes and personal aesthetics are opaque to explanation and introspection. One of the subeventual appraisals leads to what OC&C calls prospect-based emotions. What this amounts to is judging a current event based on its likely causal implications. This kind of temporal connotation is how Emotus Ponens is able to characterise events by situating it in a contextual neighborhood of concepts (Liu et al., 2003a). Emotus Ponens also bears other parallels to OC&C's theory insofar as EP parses out events, people, and things from text and makes those three semantic types the subject of appraisal. Events lead to causal projection and abductive explanation, while people and things are characterised based on their typical properties and class-based inheritances. Also, there is agreement between the two works on the tenet that surface linguistic features not dominate the assessment of emotions. The emotions contained in text amounts to more than an integration of surface features but rather requires digging down to the signified. Another parallel between WWTT and OC&C is that in WWTT, an affective valence is associated with both an event in-the-vague, and also a found root cause of an event (mapping to OC&C's agents); also, WWTT's identification of self-conscious emotions in the manner of Marvin Minsky's The Emotion Machine (forthcoming) maps well to OC&C's self-agent attributions.

Other insights in OC&C. First is concerning the intensity modulation of emotions. The authors argue that realism, self-involvedness, attention-grabbing, and prior arousal constitute an important global context for determining the intensity of an emotional reaction. There are also a slew of emotion-specific and event/agent/object -specific priors which serve as reference points against a reaction can be made. Second, in the proposed scheme, certain emotions can be characterised as compound emotions if they require the presence of some combination of event/agent/object reaction. For example, gratification and anger both require a prospect-irrelevant well-being event appraisal combined with a self-agent attribution appraisal. One final insight: this cognitive appraisal does not necessarily have to be consciously accessible.